US 7936302 B2 Abstract A method and apparatus are described for the unwrapping of a set of phase values observed for an incoming signal on a phased array antenna. The difference between values observed on adjacent elements in the array forms a first data set. The differences between adjacent ordinates in the first data set forms a second data set. The values in the second data set are rounded to the nearest whole multiple of one complete cycle before the differencing process is reversed to provide the values (representing a whole number of complete cycles) which are added to the observed phase values to provide the unwrapped phase values.
Claims(8) 1. A method of processing a signal received by a phased array antenna, said method comprising:
(i) receiving the signal via a plurality of antenna elements of said phased array antenna, said antenna elements being situated at a set of n loci;
(ii) measuring the phase of the signal at each locus to produce a set of n sequential phase values;
(iii) calculating the differences between neighboring phase values in the sequence according to:
DIFF1_{k}=Φmeasured_{k+1}−Φmeasured_{k }(k=1 to n−1) where Φmeasured_{k }is the kth phase value in the sequence;
(iv) calculating the differences between neighboring values of DIFF1_{k }according to:
DIFF2_{k}=DIFF1_{k+1}−DIFF1_{k }(k=1 to n−2) (v) rounding the values of DIFF2_{k }to the nearest integral multiple of complete phase cycles to produce the set of rounded values DIFF_{k};
(vi) summing neighboring values in the set of rounded values DIFF_{k }to provide a set of values, dΦ_{k}, according to:
dΦ _{k+1} =dΦ _{k}+Diff_{k } dΦ _{1}=0 (k=1 to n−2) (vii) summing neighboring values of dΦ_{k }to give the set of values Φ_{k }according to:
Φ_{k+1}=Φ_{k} +dΦ _{k } Φ_{0}=0 (k=1 to n−1) and
(viii) adding the values Φ_{k }to the corresponding values Φmeasured_{k }to produce unwrapped phase values.
2. The method of
3. The method of
4. The method of
5. Apparatus for processing a signal comprising:
(i) means for receiving the signal at a set of n loci,
(ii) means for measuring the phase of the signal at each locus to produce a set of n sequential phase values;
(iii) means for calculating the differences between neighboring phase values in the sequence according to:
DIFF1_{k}=Φmeasured_{k+1}−Φmeasured_{k }(k=1 to n−1) where Φmeasured_{k }is the kth phase value in the sequence;
(iv) means for calculating the differences between neighboring values of DIFF1_{k }according to:
DIFF2_{k}=DIFF1_{k+1}−DIFF1_{k }(k=1 to n−2) (v) means for rounding the values of DIFF2_{k }to the nearest integral multiple of complete phase cycles to produce the set of rounded values DIFF_{k};
(vi) means for summing neighboring values in the set of rounded values DIFF_{k }to provide a set of values, dΦ_{k}, according to:
dΦ _{k+1} =dΦ _{k}+Diff_{k}, Φ_{1}=0 (k=1 to n−2) (vii) means for summing neighboring values of dΦ_{k }to give the set of values Φ_{k }according to:
Φ_{k+1}=Φ_{k} +dΦ _{k}, Φ_{0}=0 (k=1 to n−1) and
(viii) means for adding the values Φ_{k }to the corresponding values Φmeasured_{k }to produce unwrapped phase values.
6. The apparatus of
7. The apparatus of
8. The apparatus of
Description This application is a national stage of PCT International Application No. PCT/GB2006/050315, filed Oct. 15, 2006, which claims priority under 35 U.S.C. §119 to British Patent Application Nos. 0520332.8, filed Oct. 6, 2005 and 0524624.4, filed Dec. 2, 2005, the entire disclosures of which are herein expressly incorporated by reference. The invention is concerned with the calibration of phased array antennas of the type used in applications such as Direction Finding (DF), signal separation and enhanced reception or simple beam steering. These techniques are well known but one problem commonly encountered is that knowledge is required of the response of the array to signals arriving from different directions. The set of complex responses across an array of n elements may be termed a point response vector (PRV) and the complete set of these vectors over all directions is known as the array manifold (of n dimensions). Normally a finite sampled form of the manifold is stored for use in the DF processing. The (sampled) manifold can be obtained, in principle, either by calibration or by calculation or perhaps by a combination of these. Calibration, particularly over two angle dimensions (for example azimuth and elevation) is difficult and expensive, and calculation, particularly for arrays of simple elements, is much more convenient. In this case, if the positions of the elements are known accurately (to a small fraction of a wavelength, preferably less than 1%) the relative phases of a signal arriving from a given direction can be calculated easily, at the frequency to be used. The relative amplitudes should also be known as functions of direction, particularly for simple elements, such as monopoles or loops. If the elements are all similar and oriented in the same direction then the situation corresponds to one of equal, parallel pattern elements, and the relative gains across the set of elements are all unity for all directions. The problem with calculating the array response is that this will not necessarily match the actual response for various reasons. One reason is that the signal may arrive after some degree of multipath propagation, which will distort the response. Another is that the array positions may not be specified accurately, and another that the element responses may not be as close to ideal as required. Nevertheless, in many practical systems these errors are all low enough to permit satisfactory performance to be achieved. However, one further source of error that it is important to eliminate, or reduce to a low level, is the matching of the channels between the elements and the points at which the received signals are digitized, and from which point no further significant errors can be introduced ( One solution to channel calibration is to feed an identical test signal into all the channels immediately after the elements. The relative levels and phases of these after digitization give directly the compensation (as the negative phase and reciprocal amplitude factor) which could be conveniently applied digitally to all signals before processing, when using the system ( One problem which arises during the measurement of phase angles is that of ‘unwrapping’ the measured value. The indicated value will lie within a range having a magnitude of 360° (or 2π radians) with no indication of whether the true value equals this indicated value or includes a whole number multiple of 360°/2π radians. The term ‘unwrapping’ is used in the art to describe the process of resolving such indicated values to determine the true values. According to a first aspect of the invention, a method of processing a signal comprises the steps of: (i) receiving the signal at a set of n loci; (ii) measuring the phase of the signal at each locus to produce a set of n sequential phase values; (iii) calculating the differences between neighboring phase values in the sequence according to:
(iv) calculating the differences between neighboring values of DIFF1_{k }according to:
(v) rounding the values of DIFF2_{k }to the nearest integral multiple of complete phase cycles to produce the set of rounded values DIFF_{k }; (vi) summing neighboring values in the set of rounded values DIFF_{k }to provide a set of values, dΦ_{k}, according to:
(vii) summing neighboring values of dΦ_{k }to give the set of values Φ_{k }according to:
(viii) adding the values Φ_{k }to the corresponding values Φmeasured_{k }to produce the unwrapped phase values. According to a second aspect of the invention, apparatus for processing a signal comprises: (i) means for receiving the signal at a set of n loci, (ii) means for measuring the phase of the signal at each locus to produce a set of n sequential phase values; (iii) means for calculating the differences between neighboring phase values in the sequence according to:
(iv) means for calculating the differences between neighboring values of DIFF1_{k }according to:
(v) means for rounding the values of DIFF2_{k }to the nearest integral multiple of complete phase cycles to produce the set of rounded values DIFF_{k}; (vi) means for summing neighboring values in the set of rounded values DIFF_{k }to provide a set of values, dΦ_{k}, according to:
(vii) means for summing neighboring values of dΦ_{k }to give the set of values Φ_{k }according to:
(viii) means for adding the values Φ_{k }to the corresponding values Φmeasured_{k }to produce unwrapped phase values. For any array, the phase response across the array is a funcion of the element positions. For example, for a linear array the phase response across the array is a linear function of the element positions along the axis of the array, and this is the case whatever the direction of the observed signal (though the line has different slopes for different signal directions, of course). Thus if a signal of opportunity is available the received array phases are determined and the best linear fit to these values, as related to element position, is determined. It is assumed that this linear response is close to the ideal response for this signal and that the deviations of the received values from this line are the phase errors which require compensation. In the case of equal, parallel element patterns, the amplitude responses should be equal so variations, as factors, from a mean (in this case the geometric mean) give the required corrections. Other objects, advantages and novel features of the present invention will become apparent from the following detailed description of the invention when considered in conjunction with the accompanying drawings. The following detailed description is concerned with the case of a one-dimensional antenna array having evenly spaced elements. However, this should not be seen as limiting as the invention is equally applicable to array antennas of other shapes or configuration (e.g., two dimensional planar, spherical etc), whether or not the array elements are evenly spaced (so long as the element positions are known). Referring to Here it is assumed that the relative phases have been found and that the required multiples of 2π have been added to make the phases approximately linear with element position along the array axis. This process is known as unwrapping the phase values. A number of approaches to the problem of phase unwrapping are possible and further details on how the problem may be approached are included later. Since the phase φk for each element k is directly proportional to the position xk, a plot of the (correctly adjusted) phase shifts against element positions should provide a straight line. This is the case, whatever the value of θ, the signal direction; the value of θ (and of λ) will determine the slope of the line. In practice, there will be channel phase errors which add to these path difference phases, so that the (corrected) phase values will be scattered about the line, rather than lying exactly on it ( The basis of one aspect of the invention is that, given the phase measurements and the element positions, the straight line through this set of points which gives the best fit, in some sense, is found and it is assumed that this is close to the response due to the signal. In fact it is only necessary that the slope of this line should agree with the slope due to the signal (which is 2π sin θ/λ) as any phase offset which is common to all the channels is of no physical significance. In fact if the actual signal direction is not known, then the correct slope will not be known, and the ‘best fit’ line may not have this slope exactly. However, if there is no correlation between the phase errors and the element positions, as would generally be expected to be the case, and if there is a sufficient number of elements to smooth statistical fluctuations adequately, then the match should be good. For a definition of ‘best fit’ the sum of the squares of the errors (of the given points from the line) should be minimized—i.e., a least mean square error solution is sought. Let the element positions and the phases be given by
Here With this value for b the line becomes p= The total squared error is now given by
Thus
This is the estimate of the slope of the best fit line, and putting this into the expression for e (equation (6)) gives the estimate of the channel phase matching error Channel Phase Calibration for Planar and Volume Arrays This method of the invention can be extended to apply for planar arrays and for volume, or 3D, arrays. In the planar case the phase at an element k, relative to that at the origin, is given by
The total squared error is given by
These are two simulataneous equations which can be put in the form For the volume arrays the phase of element k, again given by the inner product, is
Making the sum of the errors zero leads to
In the case of equal parallel pattern elements the gains (as real amplitude, or modulus, factors) should all be equal. If the measured gains are a_{1}, a_{2}, . . . , a_{n }then the geometric mean of these â, rather than the arithmetic means (as in the phase case) is taken, and then the error factors are a_{k}/â and the correction factors to be applied to the data before processing are the reciprocals of these. (Alternatively one could just apply factors 1/a_{k}, so effectively setting the channel gains (including the gains of the array elements) to unity. As the set of n channel outputs can be scaled arbitrarily, this is equally valid, but may require changes to any thresholds, as level sensitive quantities.) If the element patterns are not parallel (all with the same pattern shape and oriented in the same direction) then this calibration will only be valid for the direction of the signal used, which in general is not known. (Even if it is known, the calibration information could only be used for correcting the manifold vector for this single direction.) Thus this method is not applicable to mixed element arrays (e.g. containing monopoles and loops) or to arrays of similar elements (e.g. all loops) differently oriented. If the element patterns are parallel but not equal (i.e. if the array elements have different gains) then this calibration will effectively equalize all the gains, which will then agree with the stored manifold values (if this assumption has been made in computing the manifold vectors). However this will modify the channel noise levels, in the case of systems which are internal noise limited (rather than external noise limited as may be the case at HF), so that the noise is spatially ‘non-white’, which is undesirable in the processing. Thus this method is really limited to arrays with equal, parallel pattern elements, but this is in fact a very common form of array, and this calibration should be simple and effective for this ease. The method does not otherwise depend on the array geometry so is applicable to linear, planar or volume arrays. Phase Unwrapping for Regular Linear Array Considering the case of a regular linear array first, in the absence of errors the path differences between adjacent elements will all be the same, so also will be the resulting phase differences. However, the measured phases are all within an interval of 2π radians (e.g. −π to +π) so if the cumulative phase at an element is outside this range then a multiple of 2π radians will be subtracted or added, in effect, to give the observed value. In order to obtain the linear relationship between phase and element position the correct phase shifts need to be found, adding or subtracting the correct multiples of 2π to the observed values. Taking the differences between all the adjacent elements yields some that correspond to the correct phase slope, say Δφ, and some with a figure 2π higher or lower (e.g. Δφ−2π). These steps in the set of differences indicate where the increments of 2π should be added in (and to all succeeding elements). However, with channel phase errors present the difference between (Δφ+errors) and (Δφ−2π+errors) is not a simple value of 2π and it is necessary to set some thresholds to decide whether a given value is in fact near to Δφ (which itself is not known, as the signal direction is not known) or near to Δφ−2π. This problem is solved by taking a second set of differences—the differences between adjacent values of the first set. When there are two adjacent values of (Δφ+errors) their difference is (zero+errors) and when adjacent values are (Δφ+errors) and (Δφ−2π+errors) the difference is (2π+errors). Thus all the second differences are near zero, ±2π, ±4π and so on. To find the values that there would be without errors the set is simply rounded to the nearest value of 2π to get the correct, error free, second differences. (It is assumed that the errors are small enough that four such errors, some differing in sign, which accumulate in the second differences, do not reach ±π radians. An estimate of the standard deviation of the phase errors is given below, showing that up to 20° to 30° can be handled). In fact it is convenient to measure phase in cycles for this process, so that the second differences are rounded to the nearest integer. Having found the integer values for the second differences in phase (measured in cycles) the process is now reversed: starting with the first difference set to zero, the next difference is obtained by incrementing by the first of the second differences, and so on. Having obtained the (error-free) set of first differences, now containing integer values (in cycles), this process is repeated to find the set of cycles to be added and then these are applied to the measured set of phases to obtain the full (unwrapped) set of phases. The two differencing processes may be considered to be analogous to differentiation, the first reducing the linear slope to a constant value, Δφ (except for the integer cycle jumps), and the second reducing this constant to zero (where there are no jumps). Reversing the process is analogous to integration, which raises the problem of the arbitrary constant. In fact an error by one cycle (or more) may be present at the first difference stage, and integrating this contribution gives an additional slope of one phase cycle (or more) per element. However, the error estimation process described above is independent of the actual slope so the fact that the slope may be different from the true one makes no difference. A more formal analysis of the phase correction determination is given below, including the solution for the case where the array is not regular. Here the second differences, used to eliminate u, have to take into account the irregular values of d_{k }(and their first differences, Δd_{k}) so the expressions become more complicated. Phase Unwrapping for a Linear Array Uniform Linear Array (Array elements are evenly spaced). Let the full phase in channel k be given by
In order to remove φ_{0 }and also the effect of the arbitrary choice of reference point the first differences are formed, given by
To find the values of M_{k}, a summing operation (the inverse of the differencing process) is carried out twice. From (A4),
Let Δm_{a }and m_{b }be the arbitrary choices (or constants of ‘integration’) taken for ΔM_{1 }and M_{1 }respectively. Putting
The term (m_{b}−m_{1}) is a constant phase shift (over all k) and the term (k−1)(Δm_{a}−Δm_{1}) corresponds to a constant phase slope, so when the corrections M_{k }are added to φ_{k }to obtain Φ_{k }the irregular jumps m_{k }are correctly compensated for while adding an overall phase (when m_{b}≠m_{1}) and a change in slope (when Δm_{a}≠Δm_{1}). However, the phase error estimation of the invention is independent both of absolute phase and of the phase slope, so these differences do not affect the resultant estimates in any way. Non-uniform Linear Array The full phase is given by (A1) and the measured phase by (A2), but, in the case of the non-uniform linear array (A3) is replaced, for the first differences in phase, by
In this equation the quantities Δφ_{k}, Δd_{k }are known, the error differences Δε_{k }are not known but will be removed by rounding, at the appropriate point, and Δm_{k }is to be found, for each k. However u is unknown and while it is removed by taking second differences in the uniform case, this will not be the case here because, in general uΔd_{k+1 }and uΔd_{k }will differ so their difference does not disappear. Rearranging the equation gives
It is known that Δm_{k+1 }is integral, so if the errors are not too great, as before, the relation From this equation (the first ‘summation’) all the Δm_{k}, given Δm_{1 }could be found. As this is not known ΔM_{1 }is set to 0, and the set {ΔM_{k}} is found, equivalent, for the purpose of finding the best fit, to {m_{k}}, as shown in the section “Equivalence of set {M_{k}} and {m_{k}}” above. Thus with ΔM_{1}−0 the equation Note that (A20) is the equation, for the non-uniform case, equivalent to (A8) for the uniform case. Putting Δd_{k+1}=Δd_{k}, for the linear case, then (A20) becomes
Table 1 shows data derived from actual measurements using a one dimensional linear array with 10 equispaced elements. For convenience & simplicity of explanation, channel 1 is taken as the measurement reference, so that all measured phase shifts are relative to channel 1. Column 2 shows average values of measured phase relative to channel 1, calculated from a large number of acquired data (not shown). Column 3 shows the results of the first differencing process, i.e. the difference in phase between adjacent array elements. The entries in column 3 are given by subtracting the corresponding entry in column 2 from the next entry in column 2. Column 4 shows the results of the second differencing process: the entries in column 4 are given by subtracting the corresponding entry in column 3 from the next entry in column 3. Column 5 shows Diff_{k}, (k=1 to 8), the set of second difference values of Column 4, rounded to the nearest multple of 360° and expressed in cycles through subsequent division by −360°. (The negative sign is required to ensure the phase unwrap values will have the correct sense). The results in column 5 now need to be summed twice in order to obtain the phase unwrap values. The results of the first summation are given by:
The results of the first summation are shown in column 6. The second summation is given by
The results of the second summation are shown in column 7. Since, in this example, the rounded second differences were optionally divided by −360° to give the values shown in column 5, the results of the second summation shown in column 7 are now multiplied by 360° to give the amount of phase unwrapping to be associated with each channel. Thus, the entries in column 8 show the values to be added to the measured phases for each of the channels, in order to establish the actual phase shift of each channel, relative to channel 1. Simulation Results A program has been written to simulate a phase error mismatch problem using a regular linear array, at half wavelength spacing. The three input arguments are n, the number of elements, θ, the angle of the signal source, relative to the normal to the axis of the array, and the standard deviation of the channel phase errors. On running the program a set of n channel phase errors are taken from a zero mean normal distribution with the given standard deviation. These are added to the phases at the elements due to the signal, from direction θ, which give the linear phase response. As mentioned previously, it is convenient to express these phases in cycles, rather than radians or degrees. These phases are then reduced, by subtracting a number of whole cycles from each, to the range −½ to +½ (equivalent to −π to +π radians), to give the values that would be measured. This is the basic data that the channel error estimation algorithm would be provided with. The processing begins by ‘unwrapping’ the phases—restoring the cycles that have been removed from the approximately linear response. This is implemented by the process described previously, and relies on the errors being not too excessive. (The errors to the kth second difference are ε_{k}−2ε_{k+1}+ε_{k+2}, where ε_{k }is the error in channel k. The variance at the second difference level is thus 6σ^{2 }(from σ^{2}+4σ^{2}+σ^{2}) if σ^{2 }is the variance of the errors, so the standard deviation is increased √6 times. Thus for σ=30°, the s.d. of the second difference errors is about 73.5°, so ±180° corresponds to the 2.45 s.d. points, and the probability of exceeding these limits, and causing an error, is between 1% and 2%. If σ=20° errors occur at the 3.67 s.d. points, giving a probability of error of about 2×10^{−4}. This is the probability for each of the n−2 differences, not for the array as a whole.) Having obtained the full path difference phase shifts, the processing for evaluating the estimate of the slope a of the best fit line from equation (7) is applied and then the estimate of the channel errors is found from equation (6).
Table 2 shows five sets of errors for this example. The first line is the set of channel errors taken from the normal distribution with a standard deviation of 10°. The second line gives the cycles of error resulting from the unwrapping process—in this case there is no error in all ten channels. The third line gives the estimated errors across the ten channels, and the fourth is the difference between lines three and one—i.e. the errors in estimating the channel errors. Finally the fifth line removes the mean value from line three (on the basis that a common phase can be subtracted across the array) and an interesting result is observed. The residual errors increment regularly across the array—in other words they correspond to a linear response and so are due to a small error between the true response (corresponding to the signal direction of 30+) and the best fit line. This is not a failure of the method, but a result of the particular finite set of error data used, as indicated in Without information of the actual direction of the signal it is impossible to know what is the correct slope and the best that can be done is to make some best fit, in this case based on the least squared error solution. The slope of the best fit line matches that of the signal response if the phase error vector and the element position vector are orthogonal—i.e. if the phases and the positions are uncorrelated. This will not normally be exactly true for finite samples (10 in this simulation case) but would become more nearly true as the number of elements increases. However, examination of the phase slope error that has been introduced reveals that the DF error this introduces is small. In the example above the phase difference between elements after calibration by this method is 0.5°. With elements at a half wavelength apart the phase difference for a signal at δλ from broadside is 180° sin δθ, or 180° δθ, for a small angle. Thus in this case δθ= 0.5/180= 1/360 radians or about 0.16°. (The DF measurement error increases as secθ with movement to an angle θ from broadside, as the phase difference between elements between θ and θ+δθ is approximately 180° cos θδθ so in this case, δθ=0.16° secθ and if θ=60°, for example, δθ=0.32°. Finally some more examples are presented in Table 3.
In example (a) it can be seen that there is an error of one cycle per element in estimating the unwrapping phases. As this is a linear error across the array it does not affect the error estimates. In example (b) there is an error of one cycle on all the elements. As this is a constant phase error, again it does not affect the estimation of the slope of the line or the error estimates. In this case the residual errors are very small (giving a slope of 0.1° per element) but this is just a consequence of the particular set of errors chosen (and not related to the change of signal direction to 80°). Another run, with the same input arguments, gave errors of 1.8° per element. With high channel errors (from a distribution with a standard deviation of 30° in example (c)) the possibility of errors at the second difference stage occurs, and this is shown here. Here the sixth difference the error is 37.6°−2×(−75.1°)+17.5° which exceeds 180°, resulting in an extra cycle being inserted at this point (and the following points, because of the integration). This has caused the ‘corrected’ phase to be non-linear and led to errors. This result, however, was only obtained after several runs with these arguments, without this error appearing. On increasing the s.d. of the channel errors from 10° (in case (a)) to 20° (case (d)) it can be seen that the residual errors increase, from 1.7° per element to 2.2° per element. Of course, these values will vary statistically, and a proper estimate could only be obtained by taking a large number of cases. However, the residual errors can be expected to be generally proportional to the input error magnitudes, given by the standard deviation of the distribution. It can be expected that increasing the number of elements, and hence the number of points that the best fit process averages over, will reduce the residual errors. Comparison of (e) and (d) shows that the errors have fallen from (−)2.2° per element to 0.6, though again this comparison is for only one run in each case, and a large number should be carried out for firm data. Finally, In block 122, the difference between neighboring values of DIFF1_{k }is calculated according to the expression
A summing unit 124 then sums the neighboring values in the set of rounded values DIFF_{k }to provide a set of values dΦ_{k}, according to the expression
In block 125, neighboring values of dΦ_{k }are summed to yield a set of values Φ_{k }according to the expression
Finally, in a calculation unit 126, the values Φ_{k }are added to the corresponding values Φmeasured_{k }to produce unwrapped phase pulses. As shown in an alternative embodiment of the invention as illustrated in It should be noted that the invention also includes the system described above, and illustrated in The foregoing disclosure has been set forth merely to illustrate the invention and is not intended to be limiting. Since modifications of the disclosed embodiments incorporating the spirit and substance of the invention may occur to persons skilled in the art, the invention should be construed to include everything within the scope of the appended claims and equivalents thereof. Patent Citations
Non-Patent Citations
Referenced by
Classifications
Legal Events
Rotate |